Abstract 929: Transcriptomic analysis of pre-treatment tissue samples to predict clinical benefit to durvalumab in HIV-infected cancer patients

Introduction: Treatment with anti PD-1/PD-L1 antibodies has demonstrated clinical activity in different types of solid tumors, but only 20 to 30% of patients (pts) respond to these immune checkpoint inhibitors (ICIs). Therefore, predictive biomarkers of response that can assist in pt selection are urgently needed. Single biomarker expression, like PD-L1, may not provide enough information about cancer cells and the tumor microenviroment. Novel technologies, analyzing hundreds of genes at the same time, are needed to yield better predictive gene signatures of ICI response. The DURVAST trial analyzed the feasibility of durvalumab treatment in HIV-infected cancer pts, which are usually excluded from ICI clinical trials. The trial included 20 patients with different tumor types and yielded a disease control rate of 56%. Methods: Pre-ICI-treatment FFPE tumor tissue samples from 14 HIV-infected cancer pts (including 11 lung, 1 melanoma, 1 anal and 1 bladder cancer) were analyzed using the nCounter NanoString platform with the IO360 panel, including 770 genes involved in tumor biology, microenvironment and immune response. Gene expression results were correlated with clinical benefit (CB) (objective response and stable disease of more than 24 weeks by RECIST1.1 criteria), and compared to other predictive markers. Results: Exploratory analysis of pre-treatment gene expression profiles (GEPs) revealed differentially expressed genes (DEGs) between the pts with- and without CB. Panel-incorporated biological signatures related to tumor and immune activities were evaluated and some of the most DEGs (based on higher log2FC values and nominal p-values ≤0.05) were shown to be involved in cytokine and chemokine signaling. Although not significant, pts without CB tend to have lower expression of genes involved in cytokine and chemokine signaling (p = 0.097). In contrast, pts without CB tended to have a higher TGF beta signature scores (p = 0.318). When combining both signatures, we obtained an aggregated signature score that was significantly different between pts with- and without CB (p = 0.017). While the positive predictive values were the same for all tests, our signature score outperformed PD-L1 expression positivity by immunohistochemistry and PD-L1 RNA expression as predictors for clinical benefit with a two-fold higher sensitivity and negative predictive value. Conclusion: Gene expression analysis of pre-treatment tumor samples revealed distinct GEPs between HIV-infected cancer pts with- and without CB, where combined high baseline recruitment and activation of immune cells by cytokine and chemokine signaling pathways and low immunosuppressive TGF beta signaling pathways predict CB from durvalumab treatment. This predictive score outperformed other predictive markers of CB. These findings need further validation in an external and non-HIV infected pt cohort, in addition to a pt cohort treated with other ICIs. Citation Format: Jillian Wilhelmina Bracht, Maria Gonzalez-Cao, Teresa Moran, Judith Dalmau, Javier Garcia-Corbacho, Reyes Bernabe, Oscar Juan, Javier de Castro, Ana Gimenez-Capitan, Remedios Blanco, Erika Aldeguer, Sonia Rodriguez, Ana Drozdowskyj, Jordi Argilaguet, Julian Blanco, Julia Prado, Christian Brander, Jorge Carrillo, Bonaventura Clotet, Bartomeu Massuti, Mariano Provencio, Chung-Ying Huang, Clara Mayo de las Casas, Monica Garzon, Andres Felipe Cardona, Oscar Arrieta, Andreas Meyerhans, Miguel Angel Molina-Vila, Javier Martinez-Picado, Rafael Rosell. Transcriptomic analysis of pre-treatment tissue samples to predict clinical benefit to durvalumab in HIV-infected cancer patients [abstract]. In: Proceedings of the Annual Meeting of the American Association for Cancer Research 2020; 2020 Apr 27-28 and Jun 22-24. Philadelphia (PA): AACR; Cancer Res 2020;80(16 Suppl):Abstract nr 929.